MOPC-App/src/server/services/smart-assignment.ts

/**
 * Smart Assignment Scoring Service
 *
 * Calculates scores for jury/mentor-project matching based on:
 * - Tag overlap (expertise match)
 * - Bio/description match (text similarity)
 * - Workload balance
 * - Country match (mentors only)
 * - Geographic diversity penalty (prevents clustering by country)
 * - Previous round familiarity bonus (continuity across rounds)
 * - COI penalty (conflict of interest hard-block)
 *
 * Score Breakdown:
 * - Tag overlap: 0-40 points (weighted by confidence)
 * - Bio match: 0-15 points (if bio exists)
 * - Workload balance: 0-25 points
 * - Country match: 0-15 points (mentors only)
 * - Geo diversity: -15 per excess same-country assignment (threshold: 2)
 * - Previous round familiarity: +10 if reviewed in earlier round
 * - COI: juror skipped entirely if conflict declared
 */

import { prisma } from '@/lib/prisma'

// ─── Types ──────────────────────────────────────────────────────────────────

export interface ScoreBreakdown {
  tagOverlap: number
  bioMatch: number
  workloadBalance: number
  countryMatch: number
  geoDiversityPenalty: number
  previousRoundFamiliarity: number
  coiPenalty: number
}

export interface AssignmentScore {
  userId: string
  userName: string
  userEmail: string
  projectId: string
  projectTitle: string
  score: number
  breakdown: ScoreBreakdown
  reasoning: string[]
  matchingTags: string[]
}

export interface ProjectTagData {
  tagId: string
  tagName: string
  confidence: number
}

// ─── Constants ───────────────────────────────────────────────────────────────

const MAX_TAG_OVERLAP_SCORE = 40
const MAX_BIO_MATCH_SCORE = 15
const MAX_WORKLOAD_SCORE = 25
const MAX_COUNTRY_SCORE = 15
const POINTS_PER_TAG_MATCH = 8

// New scoring factors
const GEO_DIVERSITY_THRESHOLD = 2
const GEO_DIVERSITY_PENALTY_PER_EXCESS = -15
const PREVIOUS_ROUND_FAMILIARITY_BONUS = 10
// COI jurors are skipped entirely rather than penalized (effectively -Infinity)

// Common words to exclude from bio matching
const STOP_WORDS = new Set([
  'the', 'a', 'an', 'and', 'or', 'but', 'in', 'on', 'at', 'to', 'for', 'of', 'with',
  'by', 'from', 'as', 'is', 'was', 'are', 'were', 'been', 'be', 'have', 'has', 'had',
  'do', 'does', 'did', 'will', 'would', 'could', 'should', 'may', 'might', 'must',
  'that', 'which', 'who', 'whom', 'this', 'these', 'those', 'it', 'its', 'i', 'we',
  'you', 'he', 'she', 'they', 'them', 'their', 'our', 'my', 'your', 'his', 'her',
  'am', 'about', 'into', 'through', 'during', 'before', 'after', 'above', 'below',
  'between', 'under', 'again', 'further', 'then', 'once', 'here', 'there', 'when',
  'where', 'why', 'how', 'all', 'each', 'few', 'more', 'most', 'other', 'some',
  'such', 'no', 'not', 'only', 'own', 'same', 'so', 'than', 'too', 'very', 'can',
  'just', 'being', 'over', 'both', 'up', 'down', 'out', 'also', 'new', 'any',
])

// ─── Scoring Functions ───────────────────────────────────────────────────────

/**
 * Extract meaningful keywords from text
 */
function extractKeywords(text: string | null | undefined): Set<string> {
  if (!text) return new Set()

  // Tokenize, lowercase, and filter
  const words = text
    .toLowerCase()
    .replace(/[^\w\s]/g, ' ') // Remove punctuation
    .split(/\s+/)
    .filter((word) => word.length >= 3 && !STOP_WORDS.has(word))

  return new Set(words)
}

/**
 * Calculate bio match score between user bio and project description
 * Only applies if user has a bio
 */
export function calculateBioMatchScore(
  userBio: string | null | undefined,
  projectDescription: string | null | undefined
): { score: number; matchingKeywords: string[] } {
  // If no bio, return 0 (not penalized, just no bonus)
  if (!userBio || userBio.trim().length === 0) {
    return { score: 0, matchingKeywords: [] }
  }

  // If no project description, can't match
  if (!projectDescription || projectDescription.trim().length === 0) {
    return { score: 0, matchingKeywords: [] }
  }

  const bioKeywords = extractKeywords(userBio)
  const projectKeywords = extractKeywords(projectDescription)

  if (bioKeywords.size === 0 || projectKeywords.size === 0) {
    return { score: 0, matchingKeywords: [] }
  }

  // Find matching keywords
  const matchingKeywords: string[] = []
  for (const keyword of bioKeywords) {
    if (projectKeywords.has(keyword)) {
      matchingKeywords.push(keyword)
    }
  }

  if (matchingKeywords.length === 0) {
    return { score: 0, matchingKeywords: [] }
  }

  // Calculate score based on match ratio
  // Use Jaccard-like similarity: matches / (bio keywords + project keywords - matches)
  const unionSize = bioKeywords.size + projectKeywords.size - matchingKeywords.length
  const similarity = matchingKeywords.length / unionSize

  // Scale to max score (15 points)
  // A good match (20%+ overlap) should get near max
  const score = Math.min(MAX_BIO_MATCH_SCORE, Math.round(similarity * 100))

  return { score, matchingKeywords }
}

/**
 * Calculate tag overlap score between user expertise and project tags
 */
export function calculateTagOverlapScore(
  userTagNames: string[],
  projectTags: ProjectTagData[]
): { score: number; matchingTags: string[] } {
  if (projectTags.length === 0 || userTagNames.length === 0) {
    return { score: 0, matchingTags: [] }
  }

  const userTagSet = new Set(userTagNames.map((t) => t.toLowerCase()))
  const matchingTags: string[] = []
  let weightedScore = 0

  for (const pt of projectTags) {
    if (userTagSet.has(pt.tagName.toLowerCase())) {
      matchingTags.push(pt.tagName)
      // Weight by confidence - higher confidence = more points
      weightedScore += POINTS_PER_TAG_MATCH * pt.confidence
    }
  }

  // Cap at max score
  const score = Math.min(MAX_TAG_OVERLAP_SCORE, Math.round(weightedScore))
  return { score, matchingTags }
}

/**
 * Calculate workload balance score
 * Full points if under target, decreasing as over target
 */
export function calculateWorkloadScore(
  currentAssignments: number,
  targetAssignments: number,
  maxAssignments?: number | null
): number {
  // If user is at or over their personal max, return 0
  if (maxAssignments !== null && maxAssignments !== undefined) {
    if (currentAssignments >= maxAssignments) {
      return 0
    }
  }

  // If under target, full points
  if (currentAssignments < targetAssignments) {
    return MAX_WORKLOAD_SCORE
  }

  // Over target - decrease score
  const overload = currentAssignments - targetAssignments
  return Math.max(0, MAX_WORKLOAD_SCORE - overload * 5)
}

/**
 * Calculate country match score (mentors only)
 * Same country = bonus points
 */
export function calculateCountryMatchScore(
  userCountry: string | null | undefined,
  projectCountry: string | null | undefined
): number {
  if (!userCountry || !projectCountry) {
    return 0
  }

  // Normalize for comparison
  const normalizedUser = userCountry.toLowerCase().trim()
  const normalizedProject = projectCountry.toLowerCase().trim()

  if (normalizedUser === normalizedProject) {
    return MAX_COUNTRY_SCORE
  }

  return 0
}

// ─── Main Scoring Function ───────────────────────────────────────────────────

/**
 * Get smart assignment suggestions for a round
 */
export async function getSmartSuggestions(options: {
  roundId: string
  type: 'jury' | 'mentor'
  limit?: number
  aiMaxPerJudge?: number
}): Promise<AssignmentScore[]> {
  const { roundId, type, limit = 50, aiMaxPerJudge = 20 } = options

  // Get projects in round with their tags and description
  const projects = await prisma.project.findMany({
    where: {
      roundId,
      status: { not: 'REJECTED' },
    },
    select: {
      id: true,
      title: true,
      teamName: true,
      description: true,
      country: true,
      status: true,
      projectTags: {
        include: { tag: true },
      },
    },
  })

  if (projects.length === 0) {
    return []
  }

  // Get users of the appropriate role with bio for matching
  const role = type === 'jury' ? 'JURY_MEMBER' : 'MENTOR'
  const users = await prisma.user.findMany({
    where: {
      role,
      status: 'ACTIVE',
    },
    select: {
      id: true,
      name: true,
      email: true,
      bio: true,
      expertiseTags: true,
      maxAssignments: true,
      country: true,
      _count: {
        select: {
          assignments: {
            where: { roundId },
          },
        },
      },
    },
  })

  if (users.length === 0) {
    return []
  }

  // Get existing assignments to avoid duplicates
  const existingAssignments = await prisma.assignment.findMany({
    where: { roundId },
    select: { userId: true, projectId: true },
  })
  const assignedPairs = new Set(
    existingAssignments.map((a) => `${a.userId}:${a.projectId}`)
  )

  // ── Batch-query data for new scoring factors ──────────────────────────────

  // 1. Geographic diversity: per-juror country distribution for existing assignments
  const assignmentsWithCountry = await prisma.assignment.findMany({
    where: { roundId },
    select: {
      userId: true,
      project: { select: { country: true } },
    },
  })

  // Build map: userId -> { country -> count }
  const userCountryDistribution = new Map<string, Map<string, number>>()
  for (const a of assignmentsWithCountry) {
    const country = a.project.country?.toLowerCase().trim()
    if (!country) continue
    let countryMap = userCountryDistribution.get(a.userId)
    if (!countryMap) {
      countryMap = new Map()
      userCountryDistribution.set(a.userId, countryMap)
    }
    countryMap.set(country, (countryMap.get(country) || 0) + 1)
  }

  // 2. Previous round familiarity: find assignments in earlier rounds of the same program
  const currentRound = await prisma.round.findUnique({
    where: { id: roundId },
    select: { programId: true, sortOrder: true },
  })

  const previousRoundAssignmentPairs = new Set<string>()
  if (currentRound) {
    const previousAssignments = await prisma.assignment.findMany({
      where: {
        round: {
          programId: currentRound.programId,
          sortOrder: { lt: currentRound.sortOrder },
        },
      },
      select: { userId: true, projectId: true },
    })
    for (const pa of previousAssignments) {
      previousRoundAssignmentPairs.add(`${pa.userId}:${pa.projectId}`)
    }
  }

  // 3. COI declarations: all active conflicts for this round
  const coiRecords = await prisma.conflictOfInterest.findMany({
    where: {
      roundId,
      hasConflict: true,
    },
    select: { userId: true, projectId: true },
  })
  const coiPairs = new Set(
    coiRecords.map((c) => `${c.userId}:${c.projectId}`)
  )

  // ── Calculate target assignments per user ─────────────────────────────────
  const targetPerUser = Math.ceil(projects.length / users.length)

  // ── Calculate scores for all user-project pairs ───────────────────────────
  const suggestions: AssignmentScore[] = []

  for (const user of users) {
    // Skip users at AI max (they won't appear in suggestions)
    const currentCount = user._count.assignments
    if (currentCount >= aiMaxPerJudge) {
      continue
    }

    for (const project of projects) {
      // Skip if already assigned
      const pairKey = `${user.id}:${project.id}`
      if (assignedPairs.has(pairKey)) {
        continue
      }

      // COI check - skip juror entirely for this project if COI declared
      if (coiPairs.has(pairKey)) {
        continue
      }

      // Get project tags data
      const projectTags: ProjectTagData[] = project.projectTags.map((pt) => ({
        tagId: pt.tagId,
        tagName: pt.tag.name,
        confidence: pt.confidence,
      }))

      // Calculate existing scores
      const { score: tagScore, matchingTags } = calculateTagOverlapScore(
        user.expertiseTags,
        projectTags
      )

      const { score: bioScore, matchingKeywords } = calculateBioMatchScore(
        user.bio,
        project.description
      )

      const workloadScore = calculateWorkloadScore(
        currentCount,
        targetPerUser,
        user.maxAssignments
      )

      const countryScore =
        type === 'mentor'
          ? calculateCountryMatchScore(user.country, project.country)
          : 0

      // ── New scoring factors ─────────────────────────────────────────────

      // Geographic diversity penalty
      let geoDiversityPenalty = 0
      const projectCountry = project.country?.toLowerCase().trim()
      if (projectCountry) {
        const countryMap = userCountryDistribution.get(user.id)
        const sameCountryCount = countryMap?.get(projectCountry) || 0
        if (sameCountryCount >= GEO_DIVERSITY_THRESHOLD) {
          geoDiversityPenalty =
            GEO_DIVERSITY_PENALTY_PER_EXCESS *
            (sameCountryCount - GEO_DIVERSITY_THRESHOLD + 1)
        }
      }

      // Previous round familiarity bonus
      let previousRoundFamiliarity = 0
      if (previousRoundAssignmentPairs.has(pairKey)) {
        previousRoundFamiliarity = PREVIOUS_ROUND_FAMILIARITY_BONUS
      }

      const totalScore =
        tagScore +
        bioScore +
        workloadScore +
        countryScore +
        geoDiversityPenalty +
        previousRoundFamiliarity

      // Build reasoning
      const reasoning: string[] = []
      if (matchingTags.length > 0) {
        reasoning.push(`Expertise match: ${matchingTags.length} tag(s)`)
      }
      if (bioScore > 0) {
        reasoning.push(`Bio match: ${matchingKeywords.length} keyword(s)`)
      }
      if (workloadScore === MAX_WORKLOAD_SCORE) {
        reasoning.push('Available capacity')
      } else if (workloadScore > 0) {
        reasoning.push('Moderate workload')
      }
      if (countryScore > 0) {
        reasoning.push('Same country')
      }
      if (geoDiversityPenalty < 0) {
        reasoning.push(`Geo diversity penalty (${geoDiversityPenalty})`)
      }
      if (previousRoundFamiliarity > 0) {
        reasoning.push('Reviewed in previous round (+10)')
      }

      suggestions.push({
        userId: user.id,
        userName: user.name || 'Unknown',
        userEmail: user.email,
        projectId: project.id,
        projectTitle: project.title,
        score: totalScore,
        breakdown: {
          tagOverlap: tagScore,
          bioMatch: bioScore,
          workloadBalance: workloadScore,
          countryMatch: countryScore,
          geoDiversityPenalty,
          previousRoundFamiliarity,
          coiPenalty: 0, // COI jurors are skipped entirely
        },
        reasoning,
        matchingTags,
      })
    }
  }

  // Sort by score descending and limit
  return suggestions.sort((a, b) => b.score - a.score).slice(0, limit)
}

/**
 * Get mentor suggestions for a specific project
 */
export async function getMentorSuggestionsForProject(
  projectId: string,
  limit: number = 10
): Promise<AssignmentScore[]> {
  const project = await prisma.project.findUnique({
    where: { id: projectId },
    include: {
      projectTags: {
        include: { tag: true },
      },
      mentorAssignment: true,
    },
  })

  if (!project) {
    throw new Error(`Project not found: ${projectId}`)
  }

  // Get all active mentors with bio for matching
  const mentors = await prisma.user.findMany({
    where: {
      role: 'MENTOR',
      status: 'ACTIVE',
    },
    select: {
      id: true,
      name: true,
      email: true,
      bio: true,
      expertiseTags: true,
      maxAssignments: true,
      country: true,
      _count: {
        select: { mentorAssignments: true },
      },
    },
  })

  if (mentors.length === 0) {
    return []
  }

  const projectTags: ProjectTagData[] = project.projectTags.map((pt) => ({
    tagId: pt.tagId,
    tagName: pt.tag.name,
    confidence: pt.confidence,
  }))

  const targetPerMentor = 5 // Target 5 projects per mentor

  const suggestions: AssignmentScore[] = []

  for (const mentor of mentors) {
    // Skip if already assigned to this project
    if (project.mentorAssignment?.mentorId === mentor.id) {
      continue
    }

    const { score: tagScore, matchingTags } = calculateTagOverlapScore(
      mentor.expertiseTags,
      projectTags
    )

    // Bio match (only if mentor has a bio)
    const { score: bioScore, matchingKeywords } = calculateBioMatchScore(
      mentor.bio,
      project.description
    )

    const workloadScore = calculateWorkloadScore(
      mentor._count.mentorAssignments,
      targetPerMentor,
      mentor.maxAssignments
    )

    const countryScore = calculateCountryMatchScore(
      mentor.country,
      project.country
    )

    const totalScore = tagScore + bioScore + workloadScore + countryScore

    const reasoning: string[] = []
    if (matchingTags.length > 0) {
      reasoning.push(`${matchingTags.length} matching expertise tag(s)`)
    }
    if (bioScore > 0) {
      reasoning.push(`Bio match: ${matchingKeywords.length} keyword(s)`)
    }
    if (countryScore > 0) {
      reasoning.push('Same country of origin')
    }
    if (workloadScore === MAX_WORKLOAD_SCORE) {
      reasoning.push('Available capacity')
    }

    suggestions.push({
      userId: mentor.id,
      userName: mentor.name || 'Unknown',
      userEmail: mentor.email,
      projectId: project.id,
      projectTitle: project.title,
      score: totalScore,
      breakdown: {
        tagOverlap: tagScore,
        bioMatch: bioScore,
        workloadBalance: workloadScore,
        countryMatch: countryScore,
        geoDiversityPenalty: 0,
        previousRoundFamiliarity: 0,
        coiPenalty: 0,
      },
      reasoning,
      matchingTags,
    })
  }

  return suggestions.sort((a, b) => b.score - a.score).slice(0, limit)
}